Analyze a TradingView Backtest CSV with Pineify Deep Report

Pineify Backtest Deep Report analyzes a TradingView trade CSV to show metrics, rolling windows, return groupings, and bootstrap simulations. These views can help inspect a historical trade sample. They do not establish the strategy's future return or fix errors in the original backtest.

Verify the input first

Export the relevant trade list from TradingView and open Backtest Deep Report. Check the parsed trade count, dates, currency, initial capital, and a few entry/exit pairs against the original report. Partial exits, open positions, missing fields, and formatting differences need attention before interpreting aggregates.

TradingView's own report already includes risk and performance measures beyond net profit and win rate. The additional report offers another analysis workflow; it should not be presented as proof that the native platform lacks Sharpe or Sortino ratios.

A deeper historical sample can help answer a broader question, but Deep Backtesting has plan, data, and calculation limits. It does not automatically provide every market regime or remove selection bias.

Read metric definitions carefully

The current implementation calculates VaR from the lower tail of individual trade percentage returns. CVaR averages returns at or below that cutoff. These are trade-sample statistics, not a verified daily-loss forecast. A 95% label does not mean losses cannot exceed the displayed value.

Ulcer Index measures drawdown severity across the sampled equity points. Ulcer Performance Index is a separate return-relative-to-Ulcer-Index ratio. Recovery Factor relates net profit to maximum drawdown; it is not a forecast of recovery time.

A Kelly-style number is sensitive to estimated win rate and payoff assumptions. It should not be treated as a recommended live position size. Rare losses, costs, correlated trades, and estimation error can make historical estimates misleading.

Rolling windows and resampling

The default rolling statistics use consecutive windows of 20 trades. The rolling Sharpe calculation is not annualized. A weakening window is a historical observation; it does not prove that a live strategy will fail two months later.

The default Monte Carlo calculation draws 1,000 samples of the historical trade cash P&L values with replacement, each with the same number of trades. It adds sampled cash results to initial capital. This is not a model of changing position sizing, market regimes, or all possible future losses.

Drawdown percentiles describe those simulated samples rather than a guaranteed worst case. The displayed “Risk of Ruin” currently counts simulations whose final equity is zero or negative; it does not count every path that crosses a loss threshold along the way. The chart retains the first 50 simulated paths, not all 1,000.

MFE, MAE, and calendar patterns

MFE and MAE describe favorable and adverse excursions recorded for trades. A large MFE followed by a smaller realized gain can motivate an exit-rule investigation, but hindsight does not show that the peak was tradable or identifiable at the time. Likewise, MAE alone does not prove a stop was too loose.

Calendar heatmaps can reveal concentration in the sample. Before excluding a weekday or hour, check sample size, timezone, overlapping positions, and whether the pattern persists on untouched data. There is no universal fifty-trade threshold that makes a simulation statistically meaningful.

Data handling

CSV parsing and numerical report calculations run in the browser. The optional AI analysis is different: it sends summarized metrics, monthly returns, the latest rolling statistics, and Monte Carlo summaries to the analysis API. Its current payload excludes the raw trade and equity-curve arrays, but “nothing ever leaves the device” would still be inaccurate.

No documented user study supports the former claims of a 30% unrealized-profit discovery or a 15% improvement after changing exits. Use the report to formulate a test, then rerun the actual strategy and compare complete trade records under the same costs and execution assumptions.